ROCm seems to be tolerably decent, if you are willing to spend a couple hours, and, big if, if HIP supports all the various libraries you were relying on. CUDA has a huge support library, and ROCm has been missing not just the small fry stuff but a lot of the core stuff in the that library.
Long term, AI (& a lot of other interests) need to serve themselves. CUDA is excellently convenient, but long term I have a hard time imagining there being a worthwhile future for anything but Vulkan. There don't seem to be a lot of forays into writing good all-encompassing libraries in Vulkan yet, nor many more specialized AI/ML Vulkan libraries, so it feels largely like we more or less haven't started really trying.
Long term, AI (& a lot of other interests) need to serve themselves. CUDA is excellently convenient, but long term I have a hard time imagining there being a worthwhile future for anything but Vulkan. There don't seem to be a lot of forays into writing good all-encompassing libraries in Vulkan yet, nor many more specialized AI/ML Vulkan libraries, so it feels largely like we more or less haven't started really trying.